Hyper-Systolic Parallel Computing

Abstract

A new class of parallel algorithms is introduced that can achieve a complexity of O(n^3/2) with respect to the interprocessor communication, in the exact computation of systems with pairwise mutual interactions of all elements. Hitherto, conventional methods exhibit a communicational complexity of O(n^2). The amount of computation operations is not altered for the new algorithm which can be formulated as a kind of h-range problem, known from the mathematical field of Additive Number Theory. We will demonstrate the reduction in communicational expense by comparing the standard-systolic algorithm and the new algorithm on the connection machine CM5 and the CRAY T3D. The parallel method can be useful in various scientific and engineering fields like exact n-body dynamics with long range forces, polymer chains, protein folding or signal processing.

Comments: 32 pages, selfextracting uuencoded Z-compressed tex-file and ps-figures